| name | esm |
| description | ESM-2 family protein language models (ESMFold / ESM Atlas). Use when: (1) Amino-acid embeddings or MLM scores pinned to ESM-2, (2) Single-sequence structure context (ESMFold) or ESM Metagenomic Atlas via Foldseek, (3) Backbone embeddings for PLMSearch / LucaPhylo-style workflows. For ESM C / Cambrian prefer esmc. Upstream: https://github.com/facebookresearch/esm. Route via protein-language-model.
|
| license | MIT |
| category | analysis-tools |
| tags | ["ESM","ESM-2","ESMFold","PLM","protein-language-model","embeddings","HuggingFace"] |
| upstream | https://github.com/facebookresearch/esm |
| stage | function |
ESM (ESM-2 / ESMFold)
Upstream: facebookresearch/esm ·
Atlas: https://esmatlas.com
Citation
Lin, Z. et al. Evolutionary-scale prediction of atomic-level protein structure
with a language model. Science 379, 1123–1130 (2023).
https://doi.org/10.1126/science.ade2574
See also docs/references.md.
Analytical thinking
ESM-2 is the workhorse representation PLM for many metagenomic embedding /
remote-homology pipelines. Scaling enables ESMFold (single-sequence
structure) and the ESM Metagenomic Atlas (hundreds of millions of predicted
metagenomic structures — search via foldseek, not by re-folding every ORF).
| vs | Prefer ESM-2 when |
|---|
esmc | Reproducible / published ESM-2 pipelines |
saprot | Sequence-only embeddings without SA tokens |
prostt5 | AA embeddings / MLM; not AA↔3Di translation |
plmsearch / deepblast | Need the foundation model itself vs a search wrapper |
foldseek | Embedding/property tasks; use Foldseek for Atlas/AFDB hits |
Pin checkpoint (e.g. esm2_t33_650M_UR50D vs larger), truncation length, and that
Atlas/pLDDT structures remain computational. For ESM C see esmc; generative
ESM3 — cite separately if used.
How to run
Do not invent CLI flags — follow the installed package docs for the pinned tag.
Decision tree
Protein LM / structure from sequence?
├─ Pinned ESM-2 embeddings → esm
├─ ESM C / Cambrian → esmc
├─ Structure-aware AA+3Di → saprot
├─ Sequence → 3Di for Foldseek → prostt5
├─ Remote homology wrappers → plmsearch / deepblast
├─ Search Atlas / AFDB structures → foldseek
└─ DNA gLMs → genome-language-model
Related skills
protein-language-model · esmc · saprot · prostt5 · plmsearch ·
deepblast · lucaphylo · foldseek · metagenomics-llm · tool-selection
